ControlNet is the latest neural network structure that allows you to control diffusion models by adding extra conditions, a game changer for AI Image generation. It brings unprecedented levels of control to Stable Diffusion. So naturally this is the latest buzz going around in the AI art space.
I came across this awesome post by @fofrai (make sure to follow) who tested various brand logos using ControlNet and shared the results. Its amazing to see these logos come to life as a real world image generated by Stable Diffusion. In this post I want to share these wonderful results which shows the kind of control that can be achieved with ControlNet.
Here you can see the the result when Starbucks vector logo is submitted to Stable Diffusion using ControlNet, you get a pretty nice result where woman face and eyes are well defined and symmetrical.
Love how the backlighting creates this 3D looking image with depth yet it maintains the text and the bulls just like the logo.
This looks too easy the puma doing the full stretch like the logo.
Interesting how food elements like bread, tomato, onions are incorporated into the the Burger King logo using ControlNet.
Twitter bird – what came first the logo or the photography.
Taco Bell logo realised into a real looking bell with the text in nice brass colour.
In summary, ControlNet demonstrates how using AI models combined with other AI can produce some realistic results which retain so much of the guided shapes and elements that it becomes quite surreal to imagine what possibilities we would have when creating images using AI.
Here are a few useful links where you can quickly and easily experiment with ControlNet and Stable Diffusion. For more up to date links check out the Tools page.
- ControlNet Video – https://huggingface.co/spaces/fffiloni/ControlNet-Video
- GitHub Repository – https://github.com/lllyasviel/ControlNet
- ControlNet with Stable Diffusion – https://huggingface.co/spaces/hysts/ControlNet
- ControlNet with other model – https://huggingface.co/spaces/hysts/ControlNet-with-other-models